| 摘要 |
PURPOSE:To improve learning efficiency by obtaining an indirect parameter for showing a prediction value closely related to the judgement of cage assignment from the second neural net provided separately from a neural net for cage assignment. CONSTITUTION:Learning is first made about a neural net 8b, and each of direct parameter elements a1 to d1 and a2 to d2 at the time of occurrence of a new call signal are used as learning samples. Then, learning is made about a neural net 8a. For the aforesaid learning, indirect parameters S1 and S2, and t1 and t2 respectively closely related to cage assignment judgement are also inputted, in addition to the aforesaid direct parameters as input elements. Upon comple tion of learning with both nets 8a and 8b, corresponding input pattern elements a1 to d1 and a2 to d2 are inputted to the net 8b, when a new hall call signal is generated. As a result, the parameters S1 and S2 and t1 and t2 are outputted. Furthermore, S1 and S2, and t1 and t2 are inputted to the net 8a, together with the elements a1 to d1, and a2 to d2, and an assignment cage is judged based upon output signals A1 and A2 therefrom. |
